Cerebral Cortex, March 2017;27:18491862

doi:10.1093/cercor/bhw024
Advance Access Publication Date: 17 February 2016
Original Article

ORIGINAL ARTICLE

Language Ability Predicts Cortical Structure and

Covariance in Boys with Autism Spectrum Disorder
Megha Sharda1, Nicholas E.V. Foster1, Ana Tryfon1,2,
Krissy A.R. Doyle-Thomas3, Tia Ouimet1, Evdokia Anagnostou3,
Alan C. Evans2, Lonnie Zwaigenbaum5, Jason P. Lerch4,
John D. Lewis2 and Krista L. Hyde1,2 for NeuroDevNet ASD Imaging Group
1
International Laboratory for Brain Music and Sound Research (BRAMS), Universit de Montral, Montral,
Quebec, Canada H2V 2J2, 2Montreal Neurological Institute, Faculty of Medicine, McGill University, Montreal,
Quebec, Canada H3A 2B4, 3Holland Bloorview Kids Rehabilitation Hospital, 4The Hospital for Sick Children,
University of Toronto, Toronto, Ontario, Canada M5T 3H7 and 5Glenrose Rehabilitation Hospital, Edmonton,
Alberta, Canada T5G 0B7
Address correspondence to Megha Sharda, International Laboratory of Brain, Music and Sound Research (BRAMS), 1430 Boulevard Mont-Royal, Universit de
Montral, Montral, Quebec, Canada H2V 2J2. Email: megha.sharda@umontreal.ca

http://www.neurodevnet.ca/research/asd.

Abstract
There is signicant clinical heterogeneity in language and communication abilities of individuals with Autism Spectrum
Disorders (ASD). However, no consistent pathology regarding the relationship of these abilities to brain structure has emerged.
Recent developments in anatomical correlation-based approaches to map structural covariance networks (SCNs), combined
with detailed behavioral characterization, offer an alternative for studying these relationships. In this study, such an approach
was used to study the integrity of SCNs of cortical thickness and surface area associated with language and communication, in
46 high-functioning, school-age children with ASD compared with 50 matched, typically developing controls (all males) with
IQ > 75. Findings showed that there was alteration of cortical structure and disruption of fronto-temporal cortical covariance in
ASD compared with controls. Furthermore, in an analysis of a subset of ASD participants, alterations in both cortical structure
and covariance were modulated by structural language ability of the participants, but not communicative function. These
ndings indicate that structural language abilities are related to altered fronto-temporal cortical covariance in ASD, much more
than symptom severity or cognitive ability. They also support the importance of better characterizing ASD samples while
studying brain structure and for better understanding individual differences in language and communication abilities in ASD.

Key words: brain, communication, connectivity, cortical, language, MRI

Introduction pathophysiology underlying the development and evolution of

Despite diagnostic commonalities, there exists signicant het- these traits has emerged. In the current study, neuroimaging
erogeneity in the language and communication abilities of indi- methods were combined with detailed behavioral characteriza-
viduals with Autism Spectrum Disorder (ASD) (Geschwind and tion to better understand the relationship between brain struc-
Levitt 2007; Pelphrey et al. 2011). Consequently, no universal tural covariance and individual differences in language and

The Author 2016. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com
 1850 | Cerebral Cortex, 2017, Vol. 27, No. 3
communication in ASD. These approaches may be used to better
understand heterogeneity in ASD and related neurodevelopmen-
tal disorders (Lai et al. 2013; Ousley and Cermak 2013).
Language and Communication Differences in ASD
Although no longer part of the core diagnostic criteria (Volkmar
and Reichow 2013), atypical structural properties of language
(such as phonology, syntax, morphology, and semantics) have
long been considered central to ASD (Tager-Flusberg et al. 2005;
Boucher 2012). Moreover, language abilities are also strong pre-
dictors of later outcome in many individuals (Szatmari et al.
2003; Lombardo et al. 2015). At the same time, there is signicant
clinical heterogeneity in language and communication abilities
of individuals with ASD. (Tager-Flusberg et al. 2005; Boucher
2012; Taylor et al. 2014). These may range from typical onset and
development of language and communication abilities to signi-
cant difculties in speech and language, as well as complete
absence of verbal abilities in many cases (Tager-Flusberg et al.
2005). Past research has typically focused on identifying differ-
ences between ASD and typically developing (TD) comparison
groups as a basis to understand the development of these abilities.
However, no clear and consistent pathology has emerged (Amaral
et al. 2008; Amaral 2011; Ecker et al. 2015; Lainhart 2015). The wide-
spread heterogeneity of language and communication abilities,
combined with their role in abnormal early development, makes
it crucial to understand their link to brain and behavioral trajector-
ies in ASD and to consider distinct subtypes of the disorder as
components in research designs to better detect differences
(Amaral et al. 2008; Lai et al. 2014; Lombardo et al. 2015).
From a neurobiological perspective, language and communi-
cation have not been adequately distinguished. Our understand-
ing of language networks in the brain comes from a multitude of
studies encompassing language (sentence comprehension, word
reading) as well as communication (pragmatics-related functions
such as speaker intent, metaphor, and irony comprehension). In
traditional neuroimaging studies, in ASD and other developmen-
tal disorders, participants are typically matched on full-scale or
verbal intelligence quotients (IQ) rather than structural language
abilities (Kana et al. 2006; Mason et al. 2008; Mller et al. 2011).
IQ is a global measure of cognitive ability but is not sensitive to
core language and communication functions, leading to further
heterogeneity in samples and possible inconsistencies between
ndings across studies (Dennis et al. 2009). A growing body of evi-
dence indicates that various socio-communicative abilities and
higher order tasks are linked to structural language ability, even
in IQ-matched individuals with ASD (Rippon et al. 2007; Black
et al. 2009; Nadig et al. 2009). Consequently, there exist a number
of standardized neuropsychological tests that have been used to
characterize language and communication abilities as distinct
traits in children, both typically developing and those with devel-
opmental disorders (Condouris et al. 2003; Taylor et al. 2014). For
example, the Childrens Communication Checklist (CCC) and the
Test of Pragmatic Language (TOPL) have been used to measure
the pragmatic skills of individuals with ASD (Volden and Phillips
2010), and standardized tests such as the Clinical Evaluation
of Language Fundamentals (CELF-4) and the Peabody Picture
Vocabulary Test (PPVT) have been used to assess structural
language abilities in ASD (Boucher 2012). It is essential to consider
these measures in neuroimaging studies of language and commu-
nicative functioning in children with ASD to better understand the
relationships between behavioral and neural proles.
Two recent studies (Lai et al. 2014; Lombardo et al. 2015) have
highlighted the importance of including language phenotypes in
neuroimaging studies as a means of better understanding indi-
vidual differences in brain anatomy as well as predicting future
outcomes in individuals with ASD. While Lai et al. (2014) showed
the relationship between structural language and neuroanatomy
in older adults with ASD, the study by Lombardo et al. (2015) un-
derscored the importance of neural signatures of language as sig-
nicant predictors of later outcomes even in high-risk toddlers.
However, most studies continue to use measures of IQ alone, to
characterize clinical groups. There is urgent need to follow these
up with more detailed characterizations of brain and behavioral
phenotypes across the lifespan, to form clear and consistent the-
ories about the pathophysiology of ASD, especially in the context
of high variability. To address this, structural neuroimaging and
behavioral proling were combined in the present study, to better
understand the neural mechanisms underlying the emergence of
these traits in a group of school-age children with ASD.
Brain Connectivity Differences in ASD in Relation to
Language
Although ASD is a behaviorally dened disorder, there exist sig-
nicantly altered pathologies in terms of brain structure and
function between ASD and TD comparison groups (Ecker et al.
2015; Lainhart 2015). However, despite some common ndings,
there remain signicant inconsistencies in results from neuroi-
maging studies of ASD. These differences across studies are likely
due to variability in age and other sample characteristics such as
symptom severity, quality control of data, and the brain imaging
methods used (Amaral 2011; Uddin 2015), in addition to the high-
ly variable etiologies of the disorder (Geschwind and Levitt 2007;
Ellegood et al. 2014). However, 1 relatively consistent nding to
have emerged in the last few decades is atypical brain connectiv-
ity in social and language-based brain regions in ASD. For ex-
ample, altered brain connectivity between the frontal and
posterior temporal regions has been related to social language
and communication problems in ASD in numerous structural
and functional connectivity studies (Belmonte et al. 2004; Kana
et al. 2006; Di Martino et al. 2009; Mller et al. 2011; Eyler et al.
2012; Supekar et al. 2013; Maximo et al. 2014; Doyle-Thomas
et al. 2015; Sharda et al. 2015).
Although most brain connectivity research in ASD has fo-
cused on functional approaches, recent advances in structural
imaging offer new insights to study anatomical connectivity in
neurodevelopmental disorders. In particular, structural covari-
ance network (SCN) analyses have been used to highlight differ-
ences in cortical structure and network development in both
typical (Mechelli et al. 2005; Zielinski et al. 2010) and atypical po-
pulations (Zielinski et al. 2012; Bernhardt et al. 2013; Sharda et al.
2014; Valk et al. 2015). SCN analysis is a non-invasive, statistically
robust and well-replicated method that has been shown, in many
cases, to reect underlying white matter (WM) tracts as well as
functional intrinsic connectivity networks (Raznahan et al. 2011;
Alexander-Bloch, Giedd, et al. 2013; Evans 2013). Structural co-
variance is observed as inter-individual differences in regional
brain structure co-varying with other brain structures across
the population (Mechelli et al. 2005; Alexander-Bloch, Raznahan,
et al. 2013). Such anatomical covariance is believed to arise from
mutually trophic, developmental, and maturational inuences
(Mechelli et al. 2005) and has been shown to partly reect under-
lying WM connectivity as measured by diffusion studies as well
as functional networks. For example, in Lerch et al. (2006), the
authors showed that there was a high degree of structural covari-
ance between the Brocas and the Wernickes areas in the brain,
reecting the underlying arcuate fasciculus connecting them.
 Structural Covariance Networks of Language and Communication
Other studies since have replicated the existence of large-scale
intrinsically connected functional brain networks, such as the
default mode network, that can be topographically represented
as structural covariance patterns of cortical thickness (CT), cor-
tical surface area (CSA), or volume, within and across individuals
(Seeley et al. 2009). More recently, a study that compared various
measures of structural and functional connectivity between and
across species found that there was high correspondence between
cortical thickness covariance and functional connectivity metrics
in both human and rhesus macaques (Reid et al. 2015). Addition-
ally, structural co-variation has been corroborated by tract-tracing
studies in animals, and spatial patterns of structural covariance
appear to generally correspond to the spatial structure of WM
tracts (Dombrowski 2001; Mitelman et al. 2005; Gong et al. 2012).
As such, SCN analysis can provide a powerful tool to understand
ASD phenotypes in terms of cortical and morphometric changes.
It especially allows the study of individual differences in cortical
structure associated with behavior, as has been demonstrated in
previous studies (Lee et al. 2014; Voss and Zatorre 2015).
Moreover, SCNs allow one to study the nature of anatomical ab-
normalities in the context of large-scale networks rather than re-
gional differences alone and may further help to unify
inconsistencies from previous research on brain structure in ASD
(Haar et al. 2014; Zielinski et al. 2014; Osipowicz et al. 2015).
Despite their potential to inform about underlying patterns
of network relationships in ASD, studies using SCN-based
approaches have been limited. Most of these existing studies
have reported differences in SCNs underlying socio-cognitive
functions such as alexithymia (Bernhardt et al. 2013), theory of
mind (Valk et al. 2015), and processes related to socio-emotional
regulation (Zielinski et al. 2012), all of which identied large-scale
perturbations in fronto-posterior networks. However, only one
previous study investigated fronto-temporal SCNs of language
and it did so in a small sample of lower functioning children
with ASD (Sharda et al. 2014). In the present study, fronto-tem-
poral SCNs were studied in a larger sample of well-characterized,
higher functioning children with ASD as a function of their
language and communication abilities.
Study Objectives and Hypotheses
The main objective of the present research was to examine the
role of heterogeneity in ASD, by better characterizing language
and communication abilities in a group of school-age children.
This research improves on previous studies by including a sam-
ple of school-age children with ASD who are not only matched on
age and are of average IQ compared with the TD comparison
group, but also come from a sample of quality-controlled (Duch-
arme et al. 2015) data with detailed behavioral characterization.
The specic aims were as follows:
1. To characterize language and communication abilities in
children with ASD using reliable and standardized neuro-
psychological measures. It was expected that children with
ASD would have poorer language and communication abil-
ities compared with TD controls.
2. To assess the cortical morphology (in terms of CT and CSA)
and structural covariance in intrinsic networks of language
in ASD and TD. It was expected that morphological differ-
ences in brain regions involved in social and communication
functions (such as the frontal and temporal regions) would be
observed and that interregional differences in the patterns of
correlations of these particular regions would be altered in
the ASD group compared with controls.
Sharda et al. | 1851
3. To study the modulation of cortical morphology and struc-
tural covariance of fronto-temporal networks by language
and communication abilities in ASD and TD. It was expected
that in a school-age, average-IQ sample of ASD, differences in
higher order pragmatic and social communication abilities,
rather than structural language, would modulate interregio-
nal differences in SCNs of fronto-temporal circuitry.
This is the rst study, in a well-characterized and quality-con-
trolled sample of children with ASD, to assess network-level differ-
ences in the context of language and communication abilities in
ASD, rather than average group differences with an IQ-matched
TD group alone. Aims 1 and 3 were carried out in a subsample of
the ASD population for whom the measures were available. This
has implications for better understanding individual differences in
ASD and unifying heterogeneous ndings from previous research.
Materials and Methods
Participants
Two groups of children in the age range of 616 years participated
in the present study: 1) 46 boys with ASD and 2) 50 TD boys.
Participants were recruited as part of the NeuroDevNet Autism
Demonstration Project, a multisite initiative to study brain
structural and behavioral development in children with ASD
(Zwaigenbaum et al. 2011). Participants were recruited and tested
at 2 sites: 1) the Montreal Neurological Institute, in Montreal,
Canada, and 2) the Holland Bloorview Kids Rehabilitation
Hospital and the Hospital for Sick Children in Toronto.
Individuals with ASD were diagnosed by expert clinical judg-
ment (American Psychiatric Association 2000) informed by standard
diagnostic measures including the Autism Diagnostic Interview
Revised (ADI-R) (Lord et al. 1994) and the Autism Diagnostic
Observation Schedule (ADOS and ADOS-2; Lord et al. 2000). The
TD participants had no history of neurological or psychiatric illness
based on self-report and medical history questionnaires. All parti-
cipants had normal hearing. Exclusion criteria included IQ <75,
gestational age of 35 weeks or less, a medical history of neurological
disease, family history of ASD (for the TD group), or hearing
impairment. While any primary psychiatric diagnoses were an
exclusion criteria for TD, participants with ASD who had comorbid-
ity with disorders such as ADHD or anxiety (and consequently were
on medication) were not excluded from our sample due to the high
incidence of comorbidity with such disorders in ASD (Leyfer et al.
2006).
Cognitive and Behavioral Measures
Groups were matched on age (ASD: mean 12.9 years, SD 2.8; TD:
mean 12.8 years, SD 3.3). IQ was assessed using the full-2 score
on the Wechsler Abbreviated Scale of Intelligence (WASI) (Wechsler
1999) consisting of the Vocabulary and Matrices subtests (Table 1).
While IQ was in the typical range for both ASD and TD groups, the
mean IQ was signicantly lower in the ASD group. Intellectual im-
pairment is part of the ASD phenotype, with incidence estimated
at about 70% (Miller et al. 2013); the heterogeneity of IQ within
the ASD group reects the cognitive variability present in the popu-
lation and aids the generalizability of analytical results.
In addition to the WASI, a subgroup of ASD participants (n = 25)
completed 2 clinical assessments, namely, the Clinical Evaluation
of Language Fundamentals, 4th Edition (CELF-4) (Semel et al.
2003; Paslawski 2005) for assessing language ability and the Chil-
drens Communication Checklist-2 (CCC-2) (Bishop 1998) for asses-
sing communicative function. The CELF-4 is a clinical tool for the
 1852 | Cerebral Cortex, 2017, Vol. 27, No. 3
Table 1 Participant demographics
Participant demographics
Age in years(SD)
IQ (SD)
CELF-4b (SD)
CCC-2b(SD)
ADOS communication (SD)
ADOS social interaction (SD)
ADI-R communication (SD)
ADI-R social (SD)
ADI-R restricted and repetitive behaviors (SD)
pBV in mm3(SD)
Note: ASD, autism spectrum disorder; TD, typically developing; IQ measured by WASI (Wechslers Abbreviated Scale of Intelligence); SD, standard deviation; pBV, proxy
measure of brain volume; CELF-4, Clinical Evaluation of Language Fundamentals, 4th edition; CCC-2, Childrens Communication Checklist-2.
a
Participants were recruited from 2 sitesMontreal: 19 ASD, 12 TD and Toronto: 27 ASD and 38 TD.
b
CELF-4 and CCC-2 data reported for n = 25 ASD participants and the TD data for children on whom the tests were normed. ADOS = Autism Diagnostic Observation
Schedule, Communication and Social Interaction scores; ADI-R Autism Diagnostic Interview-Revised, communication, social and restricted and repetitive behavior
scores. The ADOS and ADI-R measures were available only for the ASD group.
identication, diagnosis, and evaluation of language in individuals
aged 621. The CELF-4 tests abilities in core language skills such as
word meanings (semantics), word and sentence structure (morph-
ology and syntax), as well as the recall and retrieval of spoken lan-
guage (memory). In the present study, the CELF-4 core language
standard score (henceforth, referred to as the CELF-4 score) was
used as a measure of overall structural language ability in ASD.
The CELF-4 core language standard score is a standardized and
norm-referenced score of general language ability. It is calculated
by summing the scaled scores of subtests for word structure, sen-
tence structure, recalling sentences, formulating sentences, con-
cepts and following directions, word classes, and word denitions.
The CCC-2 is a parent/caregiver administered 70-item rat-
ing scale used to measure childrens communication skills in
domains of pragmatics, syntax, morphology, and semantics. The
standard general communication composite standard score (hence-
forth, referred to as the CCC-2 score) was used in the present study
as a measure of childrens general pragmatics and communication
ability. The CCC-2 general composite score has been devised to
identify communication problems in children and is also a norm-
referenced standardized score derived from the sum of scaled
scores of all subtests including coherence, initiation, context, and
nonverbal communication. In addition, the Social Responsiveness
Scale (SRS) was evaluated for these participants as an index of
ASD symptom severity (Constantino et al. 2003). Unfortunately,
these measures were not available for the TD participants, and as
such the relation between structural covariance and language
measures in TD could not be assessed. Detailed demographics of
participants can be found in Table 1.
The present study was approved by local ethics committees
at the Montreal Neurological Institute, the Holland Bloorview
Kids Rehabilitation Hospital, and the Hospital for Sick Children.
Written informed consent was obtained from parents of all
participants, who gave their assent and were compensated for
their time. Quality control of the collected behavioral data was
performed through the LORIS database (Das et al. 2011).
Preparation for Neuroimaging
While neuroimaging techniques provide a key non-invasive tool
to study the human brain in relation to behavior, there are a
Group (all males) Signicance
ASD (n = 46)a TD (n = 50)a
12.9 (2.8) 12.8(3.3) P = 0.827
98.1 (14.9) 109.4 (11.2) P < 0.001
87.0 (24.5) 100.0 (15.0) P < 0.001
80.8 (12.2) 100.0 (15.0) P < 0.0001
3.0 (1.5)
7.2 (2.5)
15.8 (4.8)
19.1 (6.5)
6.8 (2.6)
4.8 105 (4.9 104) 4.8 105 (4.7 104) P = 0.922
number of practical and technical challenges while conduct-
ing neuroimaging experiments with children (Kotsoni et al. 2006;
Raschle et al. 2012). Some of these include problems of movement,
motivation, and alertness. Taking care to familiarize both parent
and child to the procedure often helps in maximizing outcomes
such as completion of planned scans and ensuring image quality
while minimizing discomfort on the part of participants. To ad-
dress these aims, the neuroimaging protocol included a detailed
orientation procedure for parents and children. Audio-visual
media aids and mock scanner trials were used in some cases to
motivate the participants. Breaks were taken whenever required.
Image Acquisition and Quality Control of data
High-resolution anatomical MRI scans were obtained for partici-
pants at one of 2 sites (in Montreal or Toronto) using a 3 T Siemens
Magnetom Trio TIM with a 12-channel head coil. An identical
scanning protocol was used at both sites, and both scanners
used version B17 of the Siemens software. Participants were
scanned using a sagittal T1-weighted, magnetization prepared gra-
dient-echo (MPRAGE) volume acquisition with a voxel resolution
of 1 1 1 mm3. A generalized auto-calibrating partially parallel
acquisitions (GRAPPA) acceleration factor of 2 was used.
Several stages of quality control were followed to prevent mo-
tion artifacts in the data. Images were inspected during the scan
session, and scans were repeated when artifacts were present,
whenever possible. A stringent 2-stage quality control procedure,
developed in-house, was followed. In the rst stage after data ac-
quisition, all T1 structural images were inspected in the LORIS
neuroimaging database (Das et al. 2011) and rated in detail for
quality defects, specically: motion artifact, reduced dynamic
range, variations in slice intensity, noise, ghosting, and incom-
plete brain coverage using a standardized procedure developed
in our laboratory. There was no difference between groups in
the degree of these artifacts. The scans that passed this stage
were then entered into the surface-based processing pipeline.
In the next stage, 2 independent raters (M.S. and N.E.V.F.) con-
ducted a second round of quality control after the gray matter
(GM) and WM surfaces were calculated (see section below). Both
raters had extensive experience in evaluating MRI images and
surfaces for artifacts. Using a customized procedure developed
 Structural Covariance Networks of Language and Communication
in the laboratory to assess quality of surface extraction, surfaces
were assessed for the presence of problems in gyrication, GM or
WM boundary correspondence with anatomy, or failure of the
surface extraction, resulting in a rating on a scale of 15. Surfaces
having a rating of 3.5 passed the quality control requirements,
and those with lower scores were excluded from further analyses.
Examples of surfaces with different kinds of QC failures are
illustrated in Supplementary Figure 1. Thirty percent (n = 43) of
the scans that entered this second phase of quality control
were discarded due to the presence of one or more of these
defects. Subsequently, group and site differences in the rating
scores were tested. There were no group differences (P = 0.65),
no site differences (P = 0.61), and no signicant group site
interactions (P = 0.56) on the quality control metrics of the
surface images. A total of 30% of the surfaces were evaluated
by both raters to establish interrater reliability using Cohens
Kappa (94%).
To ensure consistency in data across the 2 sites, the same
scanner model and rmware version was used, as noted above.
Both groups (ASD and TD controls) were recruited and tested at
both sites (Montreal and Toronto) so that group differences
would not be confounded with site differences. Before and
throughout data collection, human and plastic phantoms were
scanned to ensure that there were negligible differences across
sites in imaging results. Additionally, site was used as a covariate
in all statistical models.
Brain Structural Analyses
CT and CSA were estimated from the high-resolution T1 images
for all participants using CIVET 2.0 pipeline (Zijdenbos et al. 2002;
Ad-Dabbagh et al. 2006) running on the CBRAIN platform (Sherif
et al. 2014) developed at the Montreal Neurological Institute
(MNI). A 12-parameter linear transformation (Collins et al. 1995)
was used to correct MR images for signal intensity nonuniformi-
ties using the N3 algorithm (Sled et al. 1998), following which the
images were registered to a standard stereotaxic MNI space
(ICBM-152 template) (Evans et al. 1993; Collins et al. 1995; Fonov
et al. 2011). The registered images were then classied into GM,
WM, and cerebrospinal uid using an automatic tissue classica-
tion algorithm that accounted for partial volume effects (Zijden-
bos et al. 2002; Tohka et al. 2004). For each MR volume, the inner
surface of the cortical GM was extracted using the marching
cubes algorithm (Lorensen and Cline 1987). Following this, the
surface was adjusted based on the gradient at the graywhite
boundary, and the outer surface of the cortical GM was constructed
using the Constrained Laplacian-based Automated Segmentation
with Proximities (CLASP) algorithm, for each hemisphere separate-
ly, resulting in 4 surfaces of 41 962 vertices each (Kim et al. 2005).
Cortical surfaces were registered to a template to establish vertex
correspondence between subjects. Subsequently, a reverse linear
transformation to native image space was performed to estimate
cortical thickness at each of 81 924 cortical vertices (using the
t link metric representing distance between outer and inner cor-
tical surfaces) (Lerch and Evans 2005). Vertex-wise CSA was mea-
sured by calculating the area of each triangle in the middle
cortical surface, and the area assigned to any vertex was one-
third of the total area of all triangular facets adjoining it (Lyttel-
ton et al. 2009). The CT and CSA images were then each blurred
using a 20-mm surface-based diffusion blurring kernel in prepar-
ation for statistical analyses (Chung and Taylor 2004).
Automatically extracted GM and WM surfaces used for CT and
CSA were manually inspected for any segmentation irregularities.
Estimations of cortical surface structure in the insula are
Sharda et al. | 1853
unreliable because of the lack of a clear WM boundary. Likewise,
cortical structure measurements near the amygdala are unreli-
able because of the lack of a clear boundary between cortical
and non-cortical GM. For these reasons, results in these regions
must be treated with caution. To visualize the results, CT and
CSA maps were then projected onto a study-specic template
on an average mid-gray cortical surface, which was created by
averaging the T1 images across all participants.
Seed Selection for Structural Covariance Analysis
Seed-based analysis of anatomical covariance was performed
using the MACACC approach (Lerch et al. 2006) using methods de-
scribed previously (Bernhardt et al. 2013; Sharda et al. 2014; Valk
et al. 2015). MACACC measures the Pearson correlation coef-
cient, across subjects, between structural metrics (CT or CSA) at
a seed vertex and at each of the other vertices to generate a
group map of covariance. All statistical analyses were performed
using SurfStat (http://www.math.mcgill.ca/keith/surfstat/) in the
MATLAB environment (Mathworks Inc., http://www.mathworks.
com). Random eld theory was used to correct for multiple com-
parisons (Taylor and Worsley 2007; Chung et al. 2010).
Intrinsic connectivity network loci involved in speech and se-
mantic knowledge ( part of the canonical fronto-temporal circuit-
ry) were identied from previous literature (Seeley et al. 2009;
Zielinski et al. 2010). To map structural covariance networks in-
volved in language, these loci and their right hemisphere homo-
logues were used as seed regions. The seed regions of interest
(ROIs) were dened as spheres of 4 mm radius around coordi-
nates in the MNI space. They included left inferior frontal gyrus
(IFG; MNI coordinates: x = 50, y = 18, z = 7; containing Brocas
area), right IFG (x = 50, y = 18, z = 7), left superior temporal pole
(STP; x = 38, y = 10, z = 28), and right STP (x = 38, y = 10, z = 28).
Statistical Analysis
A vertex-wise general linear interaction model was used to com-
pare SCNs seeded from the ROIs between the TD and ASD groups
as in previous studies (Bernhardt et al. 2013; Sharda et al. 2014;
Valk et al. 2015). All models had age, IQ, and site as covariates
and mean thickness/surface area of seed ROI as the variable of
interest. All continuous variables (e.g., age, IQ) were mean-cen-
tered to make the interpretation of intercepts more meaningful
(Enders and Toghi 2007). For both CT and CSA, an additional
covariate term was added to the model to account for individual
differences in brain volume and/or head size, using the proxy
measure of brain volume ( pBV) method of Karama et al. (2011).
For each vertex i, a linear model was tted to calculate:
(a) Group differences in CT/CSA
Yi = 0 + 1 age + 2 site + 3 IQ + 4 pBV + 5 group,
where Y was either CT or CSA
(b) Seed-based structural covariance:
Yi = 0 + 1 age + 2 site + 3 IQ + 4 pBV + 5 Seed,
for within-group analyses
Yi = 0 + 1 age + 2 site + 3 IQ + 4 pBV + 5 group
+ 6 Seed + 7 Seed group, for between-group
analyses
(c) Modulation of SCN with language and communication mea-
sures after controlling for symptom severity
Y = 0+ 1 age + 2 site + 3 IQ + 4 pBV + 5 CELF-4
+ 6 CCC-2 + 7 SRS + 8 Seed + 9 Seed CELF-4
+ 10 Seed CCC-2, for ASD only.
 1854 | Cerebral Cortex, 2017, Vol. 27, No. 3
This design enabled the investigation of whole brain patterns
of seed-based structural covariance in each group as well as
between groups. The comparison of the models for each group
reected the covariance strength between cortical thickness of
the seed region and the cortical thickness in the rest of the cortex
across both groups. Reduced covariance between groups was
considered to reect altered patterns of anatomical develop-
ment. The modulation of covariance strength by language and
communication measures (LM) (CELF-4 or the CCC-2) in the
ASD group was also assessed by the same model (c), by assessing
the contrast of Seed LM for CELF-4 or CCC-2, respectively. A sub-
sequent analysis using 2 separate models for CELF-4 and CCC-2
modulations yielded the same results as 1 single model. Result-
ing anatomical correlation maps for all models were thresholded
at cluster level of P < 0.025 and corrected for family-wise error
(FWE) due to multiple comparisons based on the random eld
theory at P < 0.05 (2-tailed) (Taylor and Worsley 2007).
Results
Sample Characteristics
To characterize language and communication skills in ASD, be-
havioral data available from 25 ASD participants were examined.
The ASD group performed below TD norms (Table 1) on both the
language measures, including the CELF-4 (Paslawski 2005) and
the CCC-2 (Bishop 1998). Pearsons correlations revealed that
performance on the CELF-4 was best explained by cognitive abil-
ity as measured by the WASI (r = 0.75, P < 0.001), while perform-
ance on the CCC-2 was best explained by ASD symptom
severity as measured by the SRS (r = 0.76, P < 0.001). Furthermore,
the CELF-4 and the CCC-2 composite measures themselves were
not correlated (r = 0.32, P = 0.12). The only subtest on the CCC-2
that was correlated with CELF-4 was the syntax subtest (r = 0.48,
P = 0.02), further conrming that the 2 tests measured distinct as-
pects of behavior, with respect to language and communication.
Developmental language delay was only present in 4 out of the 25
children for whom detailed behavioral measures were present
and hence could not be used as a factor in subsequent analysis.
Figure 1. Group differences in CT and CSA in ASD versus TD. (a) Greater cortical thickness in ASD (vs. TD) was observed in left temporal, left angular gyrus, and bilateral
frontal regions. (b) Increase in cortical surface area in ASD > TD was observed in right posterior temporo-occipital and right dorsal motor regions. There were no group
differences in TD > ASD for CT or CSA. All maps are reported at P < 0.05, FWE-corrected.
Group Differences in CT and CSA
Surface-based measurements of CT and CSA revealed signicant
differences between ASD and TD (Fig. 1) after controlling for ef-
fects of age, IQ, site, and brain volume. The differences primarily
reect increased CT in ASD (vs. TD) in left temporal, left angular
gyrus, and bilateral frontal regions (P < 0.05, FWE-corrected).
There were no areas of decreased CT in ASD versus TD. Similarly,
ASD also showed signicant CSA increases versus TD in right
posterior temporo-occipital and superior medial regions (P < 0.05,
FWE-corrected). There were no areas of decreased CSA in ASD
versus TD.
Seed-Based Fronto-Temporal SCNs
Seed-based SCNs (for both CT and CSA) were subsequently iden-
tied using language-based ROIs in left and right IFG and left and
right STP. For CT SCNs, networks seeded from both left and right
superior temporal pole ROIs showed signicant reduction in ASD
compared with TD (Figs 2a,b and 3a,b; P < 0.05, FWE-corrected).
While the left STP networks for TD showed increased covariance
with target regions in the ipsilateral frontal areas ( Fig. 2), the net-
works for ASD showed signicantly reduced covariance in bilat-
eral frontal regions compared with the TD group (Fig. 3a). Right
STP networks for the ASD group demonstrated reduced covari-
ance with frontal regions in the right hemisphere alone, as re-
ected in Figure 3b. There were no differences between groups
in CSA SCN maps for language.
Structural covariance networks seeded from left IFG in TD cov-
ered a large expanse bilaterally, especially in frontal and temporal
areas. The extent of this SCN was relatively smaller in the ASD
group, particularly in bilateral temporal regions, though this result
did not reach statistical signicance in the group interaction ana-
lysis (P > 0.05; Fig. 2c). Networks centered on right IFG, however,
showed reduced covariance with bilateral temporal and left frontal
regions in ASD compared with TD ( Fig. 2d; P < 0.05, FWE-corrected).
These differences were especially signicant in left medial frontal
and right posterior regions, where the interregional covariance was
different across groups, also reected in the interaction plot in Fig-
ure 3c. A summary of results can be found in Table 2.
 Structural Covariance Networks of Language and Communication Sharda et al. | 1855

Figure 2. Seed-based SCNs within each group, ASD and TD. (a) SCNs seeded from left STP, (b) SCNs seeded from right STP, (c) SCNs seeded from left IFG, (d) SCNs seeded
from right IFG. Decreased covariance was found overall in ASD compared with TD for all 4 language-related SCNs (P < 0.05, FWE-corrected).

Figure 3. Seed-based language SCNs with decreased covariance in ASD versus TD. (a) Reduced covariance in left STP SCNs in ASD in bilateral frontal regions indicated in
the interaction plot below. (b) Reduced covariance between right STP and right frontal regions in ASD. (c) Reduced covariance in right IFG SCNs in left medial frontal and
right medial posterior regions in ASD. All maps are reported at P < 0.05, FWE-corrected.
 1856 | Cerebral Cortex, 2017, Vol. 27, No. 3

Table 2 Summary of results for within- and between-group SCN analysis for ASD and TD

Figure Seed Group Target regions Cluster size P value

Within-group SCN
2a Left STP TD Right frontal, temporal, parietal, and occipital lobes 34 248 <0.001
Left frontal, temporal, parietal, and occipital lobes 32 665 <0.001
ASD Left superior temporal gyrus, left orbitofrontal 18 525 <0.001
Right superior temporal gyrus, right medial superior frontal gyrus 6149 <0.001
Right inferior temporal lobe 3874 <0.001
2b Right STP TD Left frontal, temporal, and parietal lobes 25 144 <0.001
Right frontal, temporal, and parietal lobes 21 740 <0.001
Right medial frontal lobes 1586 <0.001
ASD Right inferior and superior frontal gyrus 10 164 <0.001
Right anterior temporal lobe 8916 <0.001
Left superior temporal lobe 2355 <0.001
Left orbitofrontal cortex, insula 1080 <0.001
Left posterior temporal lobe 595 0.005
Left pre- and postcentral gyrus 564 0.016
2c Left IFG TD Left frontal, temporal, parietal, and occipital lobes 33 017 <0.001
Right frontal, temporal, parietal, and occipital lobes 32 731 <0.001
ASD Left superior frontal gyrus 9338 <0.001
Left postcentral gyrus 2898 <0.001
Left inferior frontal gyrus 4058 <0.001
Right superior frontal regions 1925 <0.001
Right anterior temporal lobe 1888 <0.001
Right occipital regions 1207 <0.001
Right postcentral gyrus 1098 0.001
Left inferior temporal gyrus 887 0.005
Left anterior temporal pole 670 0.010
2d Right IFG TD Left frontal, temporal, parietal, and occipital lobes 31 559 <0.001
Right frontal, temporal, parietal, and occipital lobes 31 355 <0.001
ASD Right inferior and superior frontal regions 7251 <0.001
Left inferior frontal gyrus 2785 <0.001
Right posterior temporal lobe 477 0.0099
Between-group SCNs
3a Left STP ASD < TD Right inferior frontal gyrus, middle frontal gyrus, superior frontal gyrus 3426 <0.001
Left superior frontal gyrus, left orbitofrontal region 2452 <0.001
Right orbitofrontal regions, right anterior cingulate gyrus 755 0.002
3b Right STP ASD < TD Right inferior frontal gyrus 686 0.016
3c Right IFG ASD < TD Left anterior cingulate gyrus 947 0.012
Right lingual gyrus 581 0.048
Within-group CT and SCNs for ASD
4a, 4b Negative CELF-4 (CT) Left posterior superior temporal regions, anterior cingulate 729 0.021
Left medial superior frontal gyrus 902 0.041
Left temporal pole and inferior temporal regions 646 0.042
4c ASD_LI-ASD_LN Right inferior frontal gyrus and frontal operculum 2140 <0.001
Right temporal lobe 1644 <0.001
Left posterior superior temporal regions 942 0.004
Left medial superior frontal gyrus 815 0.015
4d CELF-4 Left IFG Left superior temporal gyrus, left orbitofrontal region 1470 0.002
Right precentral gyrus 1217 0.012
Left postcentral gyrus 1132 0.053

Modulation of Cortical Structure and Covariance by

Language Ability
Correlational analyses of CT versus language ability as measured
by CELF-4 revealed that language ability negatively modulated
cortical thickness in bilateral temporal and left medial frontal re-
gions in ASD (Fig. 4a, P < 0.001). In other words, poorer language
ability was related to greater cortical thickness in ASD as reected
in the scatter plot in Figure 4b. To further investigate this effect,
the ASD group was subdivided into language-normal (LN, CELF-4
score >90) and language-impaired subgroups (LI; CELF-4 score
< 90) based on the median split of their CELF-4 core language
standard scores. Subsequently, group differences in the CT and
CSA maps were examined. As shown in Figure 4c, the ASD-LI
group showed cortical thickening compared with the ASD-LN
group in some of the same fronto-temporal regions modulated
by CELF-4 ability ( Fig. 4a). There was no effect of CELF-4 on
CSA. Communication abilities of the ASD group as measured by
the CCC-2 were not related to either CT or CSA (P > 0.05).
To study the modulation of fronto-temporal SCNs by behav-
ioral scores of language ability, parametric terms were included
for interaction between the seed of interest and the language
measure. Results revealed that both the left STP and the left IFG
 Structural Covariance Networks of Language and Communication Sharda et al. | 1857

Figure 4. Modulation of cortical structure by language ability. (a and b) Better language abilities of ASD individuals, as measured by CELF-4, were related to reduced cortical
thickness in some of the same posterior temporal regions that had atypical cortical structure compared with TD. (c) Subgroups of ASD based on language ability reect the
same pattern of increased cortical thickness associated with poorer language abilities in the same posterior temporal regions that show atypical thickening in ASD (Refer
to Fig. 1a). (d) Positive modulation of left IFG SCN by CELF-4 language ability in bilateral motor and ipsilateral temporal target regions. There was no effect of CCC-2 on brain
structure in ASD. All maps are reported at P < 0.05, FWE-corrected.

networks were positively modulated by CELF-4 scores. For the left
STP networks, the CELF-4 scores increased covariance between
target regions in the right STP (only at a vertex level P value), re-
ecting increased interhemispheric covariance associated with
better language ability. For the left IFG networks, the CELF-4
scores positively modulated covariance in ipsilateral temporal
and bilateral motor target regions ( Fig. 4d, P < 0.05, FWE-cor-
rected). The CCC-2 score did not modulate any of the canonical
fronto-temporal SCNs.
Discussion
The present study examined and contrasted the roles of language
and communication abilities on cortical structure and covariance
of fronto-temporal networks in a sample of high-functioning,
school-age children with ASD, in comparison to a group of TD
controls. Results revealed signicant differences in both cortical
structure and covariance in ASD compared with TD. These in-
cluded increased cortical thickness as well as reduced structural
covariance especially in the fronto-temporal brain regions. Fur-
thermore, differences in both cortical structure and covariance
were modulated by the structural language abilities of the ASD
group as measured by the CELF-4 but not their communication
abilities, as measured by the CCC-2. In other words, individuals
with ASD who had poorer language abilities also had greater cor-
tical thickness in fronto-temporal brain regions. These ndings
indicate the importance of considering structural language abil-
ities, in the study of altered fronto-temporal cortical structure
and covariance in ASD. They also suggest that diagnostic speci-
ers, such as language, can be much more useful tools for
understanding heterogeneity in clinical samples, compared
with either symptom severity or cognitive ability.
Altered Cortical Thickness and Surface Area in ASD
Versus TD
Alterations in CT and CSA in ASD (vs. TD) were observed in largely
nonoverlapping regions but still involving frontal and temporal
cortices. In particular, increased CT in ASD was observed in asso-
ciation and higher order language cortices, mainly in the left
hemisphere. Increased CSA in ASD was observed in motor corti-
ces and temporo-occipital regions, mainly in the right hemi-
sphere. These ndings provide evidence for regional structural
differences in networks subserving socio-communicative and
language functioning, and are consistent with previous ndings
in ASD (Doyle-Thomas et al. 2013; Sharda et al. 2014; Zielinski
et al. 2014; Balardin et al. 2015; Foster et al. 2015; Osipowicz
et al. 2015). The atypical increase in CT and CSA in ASD is also
consistent with histopathological ndings from recent post-
mortem studies (McKavanagh et al. 2015) that showed wider
minicolumn and axonal bundle width in ASD in similar temporal
and frontal regions. A very recent study by Scholtens et al. (2015)
has demonstrated that automated measurements of cortical thick-
ness from MRI images correspond to the width of the cortical
mantle as measured by the histological measurements of von
Economo neurons in the cortex, using a cross-validation tech-
nique. It has also been speculated (Zielinski et al. 2014) that brain
regions with similar trajectory abnormalities in ASD may share
conserved embryologic, genetic, molecular, or cellular, or network
connectivity characteristics. Indeed, the trajectory of total brain
volume and cortical thickness in ASD may reect abnormal
 1858 | Cerebral Cortex, 2017, Vol. 27, No. 3
underlying cellular/molecular processes and synaptic pruning
underlying atypical connectivity. Furthermore, the nonoverlap-
ping differences in CT and CSA support the notion that these me-
trics have distinct spatial distributions and represent independent
aspects of cortical structure (Hogstrom et al. 2013; Libero et al. 2014;
Ecker et al. 2015; Foster et al. 2015). Cortical thickness, for example,
reects axon and dendrite remodeling, myelination, and synaptic
pruning in a dynamic process lasting from birth into adulthood
and is related to number of neurons in a cortical minicolumn (Pon-
tious et al. 2008; Hogstrom et al. 2013). Cortical surface area, in con-
trast, reects the process of neural stem cell proliferation and
migration early in embryologic development, reecting the num-
ber of minicolumns themselves (Rakic 1995). Thus, ndings of
early brain enlargement in ASD appear to be preferentially driven
by CSA rather than CT (Hazlett et al. 2011), suggesting the further
need to identify the specicity of atypical neuropathology in ASD
to better understand the disorder.
However, despite some consistencies, there are also differ-
ences between current and previous ndings of cortical structure
in ASD. For example, Haar et al. (2014) found no signicant struc-
tural brain differences in terms of either surface-based or vol-
ume-based metrics in a large sample (n = 1000) of ASD versus
TD individuals. In contrast, Osipowicz et al. (2015) found de-
creases in GM volume in brain regions including the thalamus,
cerebellum, anterior medial temporal lobes, and orbitofrontal re-
gions in ASD versus TD in the same sample. There are many such
contradictory ndings of increased as well as decreased cortical
thickness and GM volume-based differences, impeding the
emergence of a consistent pathophysiology of ASD (Amaral
et al. 2008; Ecker et al. 2015; Lainhart 2015). These inconsistent
ndings across ASD studies are likely due to different criteria
for quality control of images, methods used to measure brain
structure as well as subject inclusion criteria, and signal the im-
portance of controlling these factors to nd reliable differences in
cortical structure. Furthermore, the heterogeneity of phenotypes,
even in the higher functioning (IQ > 75) range of ASD, emphasize
the importance of including behavioral characteristics such as
language functions, as done here, in models of analysis. This
would not only lead to better characterization of samples, but
also to determine the factors that can best explain brain struc-
tural differences in ASD.
Disturbance of Structural Covariance Networks of
Language in ASD Versus TD
The present study revealed signicantly reduced covariance in
ASD (compared with TD) in networks seeded from fronto-tem-
poral locations. Altered SCN structure was most prominent in
the left anterior temporal lobe networks in ASD. The anterior
temporal lobe in ASD has been implicated in semantic processing
(Eyler et al. 2012), and more recently in socio-emotional regula-
tion, theory of mind processes (Olson et al. 2007), as well as lan-
guage and multimodal perceptual analysis (Bi et al. 2011).
Reduced inter- and intrahemispheric covariance was also
found in temporo-frontal regions, reecting reduced brain lat-
eralization of language networks in ASD, consistent with previ-
ous ndings from functional imaging studies (Flagg et al. 2005;
Lindell and Hudry 2013; Nielsen et al. 2014). Interhemispheric
interaction has long been a focus of developmental models for
learning and cognition, as well as genetic and neurodevelopmen-
tal disorders (Paul et al. 2007). In addition, reduced covariance of
left temporal rather than left frontal regions in high-functioning
children with ASD likely reects impairment in overall structural
language function rather than speech alone (Courchesne and
Pierce 2005; Groen et al. 2008). This is especially apparent in con-
trast to a previous study that identied fronto-temporal SCNs in
low-functioning children with ASD in the left IFG (Sharda et al
2014). Reductions in fronto-temporal SCNs in the Sharda et al.
(2014) sample were particularly governed by verbal ability (VIQ <
70), suggesting that left IFG SCNs are modulated by verbal ability
but left TP SCNs are modulated more by global language function
in a subgroup of higher functioning individuals with ASD.
Reduced SCNs have been shown to reect atypical cortical de-
velopment in terms of reduced connectivity (anatomical or func-
tional), or decoupling between specic regions (Voss and Zatorre
2015). While both CT and CSA contribute to the development of
the cortex, our ndings of reduced SCNs in ASD were only
found in terms of CT, but not in CSA. These ndings suggest
that GM differences in ASD are due to CT (rather than CSA) and
thus reect atypical dendritic arborisation, pruning, and myelin-
ation, rather than altered cell division and cell number (that are
characteristic of CSA) (Ecker et al. 2015; McKavanagh et al. 2015).
These results are consistent with previous functional and
structural studies implicating altered connectivity between
frontal and posterior regions associated with both language
and communication functions in ASD (Kana et al. 2006; Mason
et al. 2008; Anderson et al. 2011; Eyler et al. 2012; Rudie and Da-
pretto 2013; Verly et al. 2013; Sharda et al. 2015). Furthermore,
these results are particularly important, because they also distin-
guish between patterns of interregional correlations in the same
fronto-temporal network based on specic language skills. They
also suggest that the variability in neuroanatomical (and func-
tional) ndings associated with language abilities may be driven
by specic behavioral traits rather than global functional out-
comes that are measured by tests of IQ.
Modulation of Cortical Structure and Covariance by
Language Rather than Communication Ability in ASD
To examine the role of heterogeneity in language and communica-
tion abilities on cortical structure and covariance in ASD, 2 distinct
behavioral tests, the CELF-4 and the CCC-2, were administered.
Findings revealed that language abilities as measured by CELF-4
(rather than communication) negatively modulated brain structure
in regions that showed atypical pathology in ASD, specically, the
bilateral temporo-frontal circuitry, suggesting a relation between
this particular language phenotype and brain structure in ASD. Fur-
ther subdividing the ASD group based on language ability mea-
sured by the CELF-4 also illustrated similar differences in CT ( Fig.
4c). In other words, children with poorer language (ASD-LI) had
thicker cortices than children with better language (ASD-LN) in
some of the same temporo-frontal regions that showed atypical
morphology in ASD versus TD ( Fig. 1). There was, however, no re-
lationship between CCC-2 and cortical structure. The main purpose
of this median split analysis was to re-emphasize the result of the
regression analysis and show that both the difference in means as
well as the slope is signicant. This implies that the relation be-
tween language ability and cortical structure in ASD might be re-
ected in the forms of language-impaired subgroups within the
autism spectrum, further corroborating the relationship between
language ability and altered cortical structure. Furthermore, if lan-
guage is not explicitly characterized in this population, this might
lead to further inconsistencies between studies that generally look
at mean differences between groups matched on IQ but not on lan-
guage ability.
Similarly, greater cortical covariance was positively modu-
lated by better language abilities as measured by CELF-4. More
specically, better language ability as measured by CELF-4,
 Structural Covariance Networks of Language and Communication Sharda et al. | 1859

increased structural covariance between left and right temporal symptom-based subtypes of ASD to better understand brain
areas as well as between left IFG and left temporal as well as bi- network architecture. It would also be informative to study the
lateral motor regions. There were no effects of CCC-2 on cortical trajectories of different behavioral and language subtypes and
covariance. Even after removing symptom severity measured by their interaction with age, given that there are signicant devel-
SRS from the model (which partially overlaps with the construct opmental differences in both CT and CSA in ASD (Doyle-Thomas
measured by CCC-2 and is correlated with it), there was no effect et al. 2013; Zielinski et al. 2014). Finally, it will be important to rep-
of CCC-2 on cortical structure. This further conrms that lan- licate the present cross-sectional ndings in longitudinal studies
guage abilities (and not communication) may provide an add- to further examine the impact of language abilities on both brain
itional explanation for differences in cortical structure and structure and functional brain networks in the same populations.
covariance development in ASD as well as account for signicant
variability across ndings from different studies.
Contrary to our predictions, CELF-4 (and not CCC-2) modu- Conclusions
lated both cortical morphology and language network covariance
The present study examined and contrasted the roles of language
in ASD, suggesting that despite not being a core impairment,
and communication abilities on cortical structure and covariance
language is a useful metric to dene biologically meaningful
of fronto-temporal networks in a sample of high-functioning,
and clinically relevant subtypes for ASD (Ousley and Cermak
school-age children with ASD, in comparison to a group of TD con-
2013; Lai et al. 2014). These ndings further suggest that struc-
trols. The results further illustrate the importance of structural
tural language is an important factor in dening between-subject
language abilities as correlates of brain structural differences in
variability not just behaviorally, but also in the context of brain
ASD. These ndings have several important implications for better
structure, much more than either symptom severity or IQ. A
understanding the etiology as well as heterogeneity of ASD and re-
study that examined age, diagnosis, and IQ and their interactions
lated neurodevelopmental disorders. Furthermore, they also have
to understand the variations in CT in ASD found that even after
the potential for use as a biomarker to monitor treatments tar-
including all these factors, only 22% of the variance in CT in
geted toward improving language abilities in specic subgroups.
ASD was explained (Misaki et al. 2012), suggesting that there
were more profound factors affecting cortical development in
CT, which are currently unaccounted for. Our study is the rst
Supplementary Material
to show a role of structural language in predicting differences
in cortical thickness in school-age children with ASD. These nd- Supplementary material can be found at: http://www.cercor.
ings also demonstrate that matching of age and IQ across partici- oxfordjournals.org/.
pant groups is not enough in neuroimaging designs since, even
within high-functioning subgroups, there is signicant vari-
ability in ASD. As such, it might be better to maximize on the het- Funding
erogeneity of the ASD phenotype rather than limit study designs This work was sponsored by grants from NeuroDevNet (www.
to categorically matched groups. Using such dimensional ap- neurodevnet.ca) and the Canadian Institutes of Health Research
proaches in study designs to account for phenotypic heterogen- to K.H.
eity can help in better characterizing clinical populations, while
maintaining sensitivity to detect brain differences and still not
compromising the generalizability of ndings. Additionally, the Notes
relationship between variation in language skills and cortical
The authors thank the families who participated in this study.
structure may be used to identify subgroups of ASD with similar
We acknowledge the principal investigators and members of
etiology that might benet from targeted treatments to improve
the Pathways in ASD Study Team www.asdpathways.ca) and
language skills.
Simons Simplex Collection (www.sfari.org) projects, under
which certain clinical and diagnostic measures were collected.
Limitations and Future Directions We appreciate obtaining access to phenotypic data on the
Simons Foundation Autism Research Initiative (SFARI) Base.
The present ndings indicate that structural language abilities,
Approved researchers can obtain the SSC population data set
but not communicative functions, inform brain structural phe-
described in this study (http://sfari.org/resources/sfari-base) by
notypes in ASD. These ndings suggest the importance of includ-
applying at https://base.sfari.org. We thank the Data Coordinat-
ing such behavioral characteristics in the study of brain imaging
ing Centre at the Montreal Neurological Institute (A.C. Evans,
in ASD. Given that substantial heterogeneity has been a limiting
S. Das, C. Rogers, P. Kostopoulos, V. Fonov, I. Leppert), and Neuro-
factor in ASD research, one way forward might be to take into
DevNet Informatics Core for data management infrastructure.
account current and developmental language ability. However,
Conict of Interest: E.A. has received consultation fees from NOVAR-
there remain some limitations in the current study. These
TIS and Seaside Therapeutics and has an unrestricted grant from
include a modest sample size limited to males, as well as a
Sano Canada.
cross-sectional analysis in a relatively heterogeneous sample.
Additionally, assessment of language and communication abil-
ities only in the ASD group limits the generalized conclusions
about cortical structure and covariance and their relation to
References
these abilities in typical development. Furthermore, limitations Ad-Dabbagh Y, Lyttelton O, Muehlboeck JS, Lepage C, Einarson D,
of the surface-based analyses, such as vulnerabilities of the Mok K, Ivanov O, Vincent RD, Lerch J, Fombonne E, et al. 2006.
algorithm in regions such as the insula and the medial frontal The CIVET image-processing environment: a fully automated
lobe, should be kept in mind while interpreting the ndings, comprehensive pipeline for anatomical neuroimaging re-
as mentioned in the Methods section. Future studies should search. Proceedings of the 12th annual meeting of the
focus on adopting analytical approaches that incorporate organization for Human Brain Mapping. p. S45.
 1860 | Cerebral Cortex, 2017, Vol. 27, No. 3
Alexander-Bloch A, Giedd JN, Bullmore E. 2013. Imaging struc-
tural co-variance between human brain regions. Nat Rev
Neurosci. 14:322336.
Alexander-Bloch A, Raznahan A, Bullmore E, Giedd J. 2013. The
convergence of maturational change and structural covari-
ance in human cortical networks. J Neurosci. 33:28892899.
Amaral DG. 2011. The promise and the pitfalls of autism research:
an introductory note for new autism researchers. Brain Res.
1380:39.
Amaral DG, Schumann CM, Nordahl CW. 2008. Neuroanatomy of
autism. Trends Neurosci. 31:137145.
American Psychiatric Association. 2000. Diagnostic and statistical
manual of mental disorders. 4th ed. American Psychiatric
Publishing: Washington, DC.
Anderson JS, Nielsen JA, Froehlich AL, DuBray MB, Druzgal TJ,
Cariello AN, Cooperrider JR, Zielinski BA, Ravichandran C,
Fletcher PT, et al. 2011. Functional connectivity magnetic res-
onance imaging classication of autism. Brain. 134:37423754.
Balardin JB, Sato JR, Vieira G, Feng Y, Daly E, Murphy C, Murphy D,
Ecker C. 2015. Relationship between surface-based brain mor-
phometric measures and intelligence in autism spectrum dis-
orders: inuence of history of language delay. Autism Res. doi:
10.1002/aur.1470.
Belmonte MK, Allen G, Beckel-Mitchener A, Boulanger LM,
Carper RA, Webb SJ. 2004. Autism and abnormal development
of brain connectivity. J Neurosci. 24:92289231.
Bernhardt BC, Valk SL, Silani G, Bird G, Frith U, Singer T. 2013. Se-
lective disruption of sociocognitive structural brain networks
in autism and alexithymia. Cereb Cortex. 24:32583267.
Bi Y, Wei T, Wu C, Han Z, Jiang T, Caramazza A. 2011. The role of
the left anterior temporal lobe in language processing revis-
ited: evidence from an individual with ATL resection.
Cortex. 47:575587.
Bishop DVM. 1998. Development of the Childrens Communica-
tion Checklist (CCC): a method for assessing qualitative as-
pects of communicative impairment in children. J Child
Psychol Psychiatry. 39:879891.
Black DO, Wallace GL, Sokoloff JL, Kenworthy L. 2009. Brief report:
IQ split predicts social symptoms and communication abil-
ities in high-functioning children with autism spectrum dis-
orders. J Autism Dev Disord. 39:16131619.
Boucher J. 2012. Research review: structural language in autistic
spectrum disorder - characteristics and causes. J Child
Psychol Psychiatry. 53:219233.
Chung MK, Taylor J. 2004. Diffusion smoothing on brain surface
via nite element method. IEEE International Symposium
on Biomedical Imaging: Nano to Macro. Washington, DC,
USA: IEEE. 432435 p.
Chung MK, Worsley KJ, Nacewicz BM, Dalton KM, Davidson RJ.
2010. General multivariate linear modeling of surface shapes
using SurfStat. Neuroimage. 53:491505.
Collins DL, Neelin P, Peters TM, Evans AC. 1995. Automatic 3D in-
tersubject registration of MR volumetric data in standardized
Talairach space. J Comput Assist Tomogr. 18:192205.
Condouris K, Meyer E, Tager-Flusberg H. 2003. The relationship
between standardized measures of language and measures
of spontaneous speech in children with autism. Am J
Speech Lang Pathol. 12:349358.
Constantino JN, Davis SA, Todd RD, Schindler MK, Gross MM,
Brophy SL, Metzger LM, Shoushtari CS, Splinter R, Reich W.
2003. Validation of a brief quantitative measure of autistic
traits: comparison of the social responsiveness scale with
the autism diagnostic interview-revised. J Autism Dev
Disord. 33:427433.
Courchesne E, Pierce K. 2005. Why the frontal cortex in autism
might be talking only to itself: local over-connectivity but
long-distance disconnection. Curr Opin Neurobiol. 15:225230.
Das S, Zijdenbos AP, Harlap J, Vins D, Evans AC. 2011. LORIS: a
web-based data management system for multi-center stud-
ies. Front Neuroinform. 5:37.
Dennis M, Francis DJ, Cirino PT, Schachar R, Barnes MA,
Fletcher JM. 2009. Why IQ is not a covariate in cognitive stud-
ies of neurodevelopmental disorders. J Int Neuropsychol Soc.
15:331343.
Di Martino A, Ross K, Uddin LQ, Sklar AB, Castellanos FX,
Milham MP. 2009. Functional brain correlates of social and non-
social processes in autism spectrum disorders: an activation
likelihood estimation meta-analysis. Biol Psychiatry. 65:6374.
Dombrowski SM. 2001. Quantitative architecture distinguishes
prefrontal cortical systems in the rhesus monkey. Cereb
Cortex. 11:975988.
Doyle-Thomas KAR, Duerden EG, Taylor MJ, Lerch JP, Soorya LV,
Wang AT, Fan J, Hollander E, Anagnostou E. 2013. Effects of
age and symptomatology on cortical thickness in autism
spectrum disorders. Res Autism Spectr Disord. 7:141150.
Doyle-Thomas KAR, Lee W, Foster NE V, Tryfon A, Ouimet T,
Hyde KL, Evans AC, Lewis J, Zwaigenbaum L, Anagnostou E.
2015. Atypical functional brain connectivity during rest in
autism spectrum disorders. Ann Neurol. 77:866876.
Ducharme S, Albaugh MD, Nguyen T-V, Hudziak JJ, Mateos-
Prez JM, Labbe A, Evans AC, Karama S. 2015. Trajectories of
cortical thickness maturation in normal brain development -
The importance of quality control procedures. Neuroimage.
125:267279.
Ecker C, Bookheimer SY, Murphy DGM. 2015. Neuroimaging in
autism spectrum disorder: brain structure and function
across the lifespan. Lancet Neurol. 14:11211134.
Ellegood J, Markx S, Lerch JP, Steadman PE, Gen C, Provenzano F,
Kushner SA, Henkelman RM, Karayiorgou M, Gogos JA. 2014.
Neuroanatomical phenotypes in a mouse model of the
22q11.2 microdeletion. Mol Psychiatry. 19:99107.
Enders CK, Toghi D. 2007. Centering predictor variables in cross-
sectional multilevel models: a new look at an old issue.
Psychol Methods. 12:121138.
Evans A, Collins D, Millst S, Brown E, Kelly R, Peters T. 1993. 3D stat-
istical neuroanatomical models from 305 MRI volumes. Nuclear
Science Symposium and Medical Imaging Conference; IEEE
Conference Record. San Francisco, CA: IEEE. 18131817 p.
Evans AC. 2013. Networks of anatomical covariance. Neuroimage.
80:489504.
Eyler LT, Pierce K, Courchesne E. 2012. A failure of left temporal
cortex to specialize for language is an early emerging and fun-
damental property of autism. Brain. 135:949960.
Flagg EJ, Cardy JEO, Roberts W, Roberts TPL. 2005. Language lat-
eralization development in children with autism: insights
from the late eld magnetoencephalogram. Neurosci Lett.
386:8287.
Fonov V, Evans AC, Botteron K, Almli CR, McKinstry RC,
Collins DL. 2011. Unbiased average age-appropriate atlases
for pediatric studies. Neuroimage. 54:313327.
Foster NEV, Doyle-Thomas KAR, Tryfon A, Ouimet T,
Anagnostou E, Evans AC, Zwaigenbaum L, Lerch JP, Lewis JD,
Hyde KL. 2015. Structural gray matter differences during
childhood development in autism spectrum disorder: a
multi-metric approach. Pediatr Neurol. 53:350359.
Geschwind DH, Levitt P. 2007. Autism spectrum disorders: devel-
opmental disconnection syndromes. Curr Opin Neurobiol.
17:103111.
 Structural Covariance Networks of Language and Communication
Gong G, He Y, Chen ZJ, Evans AC. 2012. Convergence and diver-
gence of thickness correlations with diffusion connections
across the human cerebral cortex. Neuroimage. 59:12391248.
Groen WB, Zwiers MP, van der Gaag R-J, Buitelaar JK. 2008. The
phenotype and neural correlates of language in autism: an in-
tegrative review. Neurosci Biobehav Rev. 32:14161425.
Haar S, Berman S, Behrmann M, Dinstein I. 2014. Anatomical ab-
normalities in autism? Cereb Cortex. 26:14401452.
Hazlett HC, Poe MD, Gerig G, Styner M, Chappell C, Smith RG,
Vachet C, Piven J. 2011. Early brain overgrowth in autism asso-
ciated with an increase in cortical surface area before age 2
years. Arch Gen Psychiatry. 68:467476.
Hogstrom LJ, Westlye LT, Walhovd KB, Fjell AM. 2013. The struc-
ture of the cerebral cortex across adult life: age-related pat-
terns of surface area, thickness, and gyrication. Cereb
Cortex. 23:25212530.
Kana RK, Keller TA, Cherkassky VL, Minshew NJ, Just MA. 2006.
Sentence comprehension in autism: thinking in pictures
with decreased functional connectivity. Brain. 129:24842493.
Karama S, Colom R, Johnson W, Deary IJ, Haier R, Waber DP,
Lepage C, Ganjavi H, Jung R, Evans AC. 2011. Cortical thickness
correlates of specic cognitive performance accounted for by
the general factor of intelligence in healthy children aged 6 to
18. Neuroimage. 55:14431453.
Kim JS, Singh V, Lee JK, Lerch J, Ad-Dabbagh Y, MacDonald D,
Lee JM, Kim SI, Evans AC. 2005. Automated 3-D extraction
and evaluation of the inner and outer cortical surfaces using
a Laplacian map and partial volume effect classication.
Neuroimage. 27:210221.
Kotsoni E, Byrd D, Casey BJ. 2006. Special considerations for func-
tional magnetic resonance imaging of pediatric populations. J
Magn Reson Imaging. 23:877886.
Lai M-C, Lombardo MV, Chakrabarti B, Baron-Cohen S. 2013. Sub-
grouping the autism spectrum: reections on DSM-5. PLoS
Biol. 11:e1001544.
Lai M-C, Lombardo MV, Ecker C, Chakrabarti B, Suckling J,
Bullmore ET, Happe F, Murphy DGM, Baron-Cohen S. 2014.
Neuroanatomy of individual differences in language in adult
males with autism. Cereb Cortex. 25:36133628.
Lainhart JE. 2015. Brain imaging research in autism spectrum dis-
orders: in search of neuropathology and health across the life-
span. Curr Opin Psychiatry. 28:7682.
Lee NR, Raznahan A, Wallace GL, Alexander-Bloch A, Clasen LS,
Lerch JP, Giedd JN. 2014. Anatomical coupling among
distributed cortical regions in youth varies as a function of in-
dividual differences in vocabulary abilities. Hum Brain Mapp.
35:18851895.
Lerch JP, Evans AC. 2005. Cortical thickness analysis examined
through power analysis and a population simulation.
Neuroimage. 24:163173.
Lerch JP, Worsley K, Shaw WP, Greenstein DK, Lenroot RK, Giedd J,
Evans AC. 2006. Mapping anatomical correlations across cere-
bral cortex (MACACC) using cortical thickness from MRI.
Neuroimage. 31:9931003.
Leyfer OT, Folstein SE, Bacalman S, Davis NO, Dinh E, Morgan J,
Tager-Flusberg H, Lainhart JE. 2006. Comorbid psychiatric dis-
orders in children with autism: interview development and
rates of disorders. J Autism Dev Disord. 36:849861.
Libero LE, DeRamus TP, Deshpande HD, Kana RK. 2014. Surface-
based morphometry of the cortical architecture of autism
spectrum disorders: volume, thickness, area, and gyrication.
Neuropsychologia. 62:110.
Sharda et al. | 1861
Lindell AK, Hudry K. 2013. Atypicalities in cortical structure,
handedness, and functional lateralization for language in
autism spectrum disorders. Neuropsychol Rev. 23:257270.
Lombardo MV, Pierce K, Eyler LT, Carter Barnes C, Ahrens-
Barbeau C, Solso S, Campbell K, Courchesne E. 2015. Different
functional neural substrates for good and poor language out-
come in autism. Neuron. 86:567577.
Lord C, Risi S, Lambrecht L, Cook EH, Leventhal BL, DiLavore PC,
Pickles A, Rutter M. 2000. The autism diagnostic observation
schedule-generic: a standard measure of social and commu-
nication decits associated with the spectrum of autism. J
Autism Dev Disord. 30:205223.
Lord C, Rutter M, Le Couteur A. 1994. Autism Diagnostic Inter-
view-Revised: a revised version of a diagnostic interview for
caregivers of individuals with possible pervasive develop-
mental disorders. J Autism Dev Disord. 24:659685.
Lorensen WE, Cline HE. 1987. Marching cubes: a high resolution 3D
surface construction algorithm. Proceedings of the 14th An-
nual Conference on Computer Graphics and Interactive Tech-
niques. SIGGRAPH 87. New York, NY, USA: ACM. 163169 p.
Lyttelton OC, Karama S, Ad-Dabbagh Y, Zatorre RJ, Carbonell F,
Worsley K, Evans AC. 2009. Positional and surface area asym-
metry of the human cerebral cortex. Neuroimage. 46:895903.
Mason RA, Williams DL, Kana RK, Minshew N, Just MA. 2008. The-
ory of Mind disruption and recruitment of the right hemisphere
during narrative comprehension in autism. Neuropsychologia.
46:269280.
Maximo JO, Cadena EJ, Kana RK. 2014. The implications of brain
connectivity in the neuropsychology of autism. Neuropsychol
Rev. 24:1631.
McKavanagh R, Buckley E, Chance SA. 2015. Wider minicolumns
in autism: a neural basis for altered processing? Brain. 138:
20342045.
Mechelli A, Friston KJ, Frackowiak RS, Price CJ. 2005. Structural co-
variance in the human cortex. J Neurosci. 25:83038310.
Miller JS, Bilder D, Farley M, Coon H, Pinborough-Zimmerman J,
Jenson W, Rice CE, Fombonne E, Pingree CB, Ritvo E, et al.
2013. Autism spectrum disorder reclassied: a second look
at the 1980s Utah/UCLA Autism Epidemiologic Study. J
Autism Dev Disord. 43:200210.
Misaki M, Wallace GL, Dankner N, Martin A, Bandettini PA. 2012.
Characteristic cortical thickness patterns in adolescents with
autism spectrum disorders: interactions with age and intel-
lectual ability revealed by canonical correlation analysis.
Neuroimage. 60:18901901.
Mitelman SA, Buchsbaum MS, Brickman AM, Shihabuddin L.
2005. Cortical intercorrelations of frontal area volumes in
schizophrenia. Neuroimage. 27:753770.
Mller R-A, Shih P, Keehn B, Deyoe JR, Leyden KM, Shukla DK.
2011. Underconnected, but how? A survey of functional con-
nectivity MRI studies in autism spectrum disorders. Cereb
Cortex. 21:22332243.
Nadig A, Vivanti G, Ozonoff S. 2009. Adaptation of object descrip-
tions to a partner under increasing communicative demands:
a comparison of children with and without autism. Autism
Res. 2:334347.
Nielsen JA, Zielinski BA, Fletcher PT, Alexander AL, Lange N,
Bigler ED, Lainhart JE, Anderson JS. 2014. Abnormal lateraliza-
tion of functional connectivity between language and default
mode regions in autism. Mol Autism. 5:8.
Olson IR, Plotzker A, Ezzyat Y. 2007. The Enigmatic temporal pole:
a review of ndings on social and emotional processing. Brain.
130:17181731.
 1862 | Cerebral Cortex, 2017, Vol. 27, No. 3
Osipowicz K, Bosenbark DD, Patrick KE. 2015. Cortical changes
across the autism lifespan. Autism Res. 8:379385.
Ousley O, Cermak T. 2013. Autism spectrum disorder: dening di-
mensions and subgroups. Curr Dev Disord Reports. 1:2028.
Paslawski T. 2005. The clinical evaluation of language fundamentals,
fourth edition (CELF-4): a review. Can J Sch Psychol. 20:129134.
Paul LK, Brown WS, Adolphs R, Tyszka JM, Richards LJ,
Mukherjee P, Sherr EH. 2007. Agenesis of the corpus callosum:
genetic, developmental and functional aspects of connectiv-
ity. Nat Rev Neurosci. 8:287299.
Pelphrey KA, Shultz S, Hudac CM, Vander Wyk BC. 2011. Research
review: constraining heterogeneity: the social brain and its
development in autism spectrum disorder. J Child Psychol
Psychiatry. 52:631644.
Pontious A, Kowalczyk T, Englund C, Hevner RF. 2008. Role of
intermediate progenitor cells in cerebral cortex development.
Dev Neurosci. 30:2432.
Rakic P. 1995. A small step for the cell, a giant leap for mankind: a
hypothesis of neocortical expansion during evolution. Trends
Neurosci. 18:383388.
Raschle N, Zuk J, Ortiz-Mantilla S, Sliva DD, Franceschi A,
Grant PE, Benasich AA, Gaab N. 2012. Pediatric neuroimaging
in early childhood and infancy: challenges and practical
guidelines. Ann N Y Acad Sci. 1252:4350.
Raznahan A, Lerch JP, Lee N, Greenstein D, Wallace GL,
Stockman M, Clasen L, Shaw PW, Giedd JN. 2011. Patterns of
coordinated anatomical change in human cortical develop-
ment: a longitudinal neuroimaging study of maturational
coupling. Neuron. 72:873884.
Reid AT, Lewis J, Bezgin G, Khundrakpam B, Eickhoff SB,
McIntosh AR, Bellec P, Evans AC. 2015. A cross-modal, cross-
species comparison of connectivity measures in the primate
brain. Neuroimage. 125:311331.
Rippon G, Brock J, Brown C, Boucher J. 2007. Disordered connect-
ivity in the autistic brain: challenges for the new psycho-
physiology. Int J Psychophysiol. 63:164172.
Rudie JD, Dapretto M. 2013. Convergent evidence of brain over-
connectivity in children with autism? Cell Rep. 5:565566.
Scholtens LH, de Reus MA, van den Heuvel MP. 2015. Linking con-
temporary high resolution magnetic resonance imaging to
the von Economo legacy: a study on the comparison of MRI
cortical thickness and histological measurements of cortical
structure. Hum Brain Mapp. 36:30383046.
Seeley WW, Crawford RK, Zhou J, Miller BL, Greicius MD. 2009.
Neurodegenerative diseases target large-scale human brain
networks. Neuron. 62:4252.
Semel E, Wiig EH, Secord WA. 2003. Clinical evaluation of language
fundamentals, fourth edition (CELF-4). Toronto, Canada: The
Psychological Corporation/A Harcourt Assessment Company.
Sharda M, Khundrakpam BS, Evans AC, Singh NC. 2014. Disrup-
tion of structural covariance networks for language in autism
is modulated by verbal ability. Brain Struct Funct. [Epub ahead
of print].
Sharda M, Midha R, Malik S, Mukerji S, Singh NC. 2015. Fronto-
temporal connectivity is preserved during sung but not spo-
ken word listening, across the autism spectrum. Autism Res.
8:174186.
Sherif T, Rioux P, Rousseau M-E, Kassis N, Beck N, Adalat R, Das S,
Glatard T, Evans AC. 2014. CBRAIN: a web-based, distributed
computing platform for collaborative neuroimaging research.
Front Neuroinform. 8:54.
Sled JG, Zijdenbos AP, Evans AC. 1998. A nonparametric method
for automatic correction of intensity nonuniformity in MRI
data. IEEE Trans Med Imaging. 17:8797.
Supekar K, Uddin LQ, Khouzam A, Phillips J, Gaillard WD,
Kenworthy LE, Yerys BE, Vaidya CJ, Menon V. 2013. Brain hy-
perconnectivity in children with autism and its links to social
decits. Cell Rep. 5:738747.
Szatmari P, Bryson SE, Boyle MH, Streiner DL, Duku E. 2003.
Predictors of outcome among high functioning children with
autism and Asperger syndrome. J Child Psychol Psychiatry.
44:520528.
Tager-Flusberg H, Paul R, Lord C. 2005. Language and communi-
cation in autism. In: Volkmar F, Paul R, Klin A, Cohen D, edi-
tors. Handbook of autism and pervasive developmental
disorders. 3rd ed. Hoboken (NJ). p. 335364.
Taylor JE, Worsley KJ. 2007. Detecting Sparse signals in random
elds, with an application to brain mapping. J Am Stat
Assoc. 102:913928.
Taylor MJ, Charman T, Robinson EB, Hayiou-Thomas ME,
Happ F, Dale PS, Ronald A. 2014. Language and traits of aut-
ism spectrum conditions: Evidence of limited phenotypic and
etiological overlap. Am J Med Genet B Neuropsychiatr Genet.
165:587595.
Tohka J, Zijdenbos A, Evans A. 2004. Fast and robust parameter
estimation for statistical partial volume models in brain
MRI. Neuroimage. 23:8497.
Uddin LQ. 2015. Idiosyncratic connectivity in autism: develop-
mental and anatomical considerations. Trends Neurosci.
38:261263.
Valk SL, Di Martino A, Milham MP, Bernhardt BC. 2015. Multicen-
ter mapping of structural network alterations in autism. Hum
Brain Mapp. 36:23642373.
Verly M, Verhoeven J, Zink I, Mantini D, Van Oudenhove L,
Lagae L, Sunaert S, Rommel N. 2013. Structural and functional
underconnectivity as a negative predictor for language in aut-
ism. Hum Brain Mapp. 35:36023615.
Volden J, Phillips L. 2010. Measuring pragmatic language in
speakers with autism spectrum disorders: Comparing the
childrens communication checklist2 and the test of prag-
matic language. Am J Speech Lang Pathol. 19:204212.
Volkmar FR, Reichow B. 2013. Autism in DSM-5: progress and
challenges. Mol Autism. 4:13.
Voss P, Zatorre RJ. 2015. Early visual deprivation changes corti-
cal anatomical covariance in dorsal-stream structures.
Neuroimage. 108:194202.
Wechsler D. 1999. Wechsler Abbreviated Scale of Intelligence.
New York (NY): The Psychological Corporation, Harcourt
Brace & Company. - Open Access Library [WWW Document].
http://www.oalib.com/references/10668025
Zielinski BA, Anderson JS, Froehlich AL, Prigge MBD, Nielsen JA,
Cooperrider JR, Cariello AN, Fletcher PT, Alexander AL,
Lange N, et al. 2012. scMRI reveals large-scale brain network
abnormalities in autism. PLoS ONE. 7:e49172.
Zielinski BA, Gennatas ED, Zhou J, Seeley WW. 2010. Network-
level structural covariance in the developing brain. Proc Natl
Acad Sci USA. 107:1819118196.
Zielinski BA, Prigge MBD, Nielsen JA, Froehlich AL, Abildskov TJ,
Anderson JS, Fletcher PT, Zygmunt KM, Travers BG, Lange N,
et al. 2014. Longitudinal changes in cortical thickness in aut-
ism and typical development. Brain. 137:17991812.
Zijdenbos AP, Forghani R, Evans AC. 2002. Automatic pipeline
analysis of 3-D MRI data for clinical trials: application to mul-
tiple sclerosis. IEEE Trans Med Imaging. 21:12801291.
Zwaigenbaum L, Scherer S, Szatmari P, Fombonne E, Bryson SE,
Hyde K, Anognostou E, Al E. 2011. The NeuroDevNet Autism
Spectrum Disorders Demonstration Project. Semin Pediatr
Neurol. 18:4048.